Wavelet-Domain Multiview Active Learning for Spatial-Spectral Hyperspectral Image Classification

Abstract

Active learning (AL) has been shown to be effective for strategic selection of training samples to support classification of hyperspectral imagery. It is well understood that the performance of classification can further be improved by utilizing the spatial information in hyperspectral images. In this paper, we propose a new wavelet-based multiview AL approach for hyperspectral image classification. Specifically, a three-dimensional redundant wavelet transform (3D-RDWT) is used to generate multiple views that are then integrated in a multiview AL framework. The spatial features generated via 3D-RDWT not only provide sufficient views for multiview AL, but are also less sensitive to additive noise. Within this framework, we also propose new query criteria that result in effective AL. An intersection-based query criterion is proposed to reduce the redundancy within the contention pool. A singularity-based criterion is also used to identify informative pixels by taking spatial information into account when selecting samples. The proposed method is evaluated on four real-world hyperspectral datasets, and the experimental results demonstrate the efficacy of the proposed method compared with traditional AL methods.